Back to Project List
ActiveAIResearchDevelopment

Physical AI Model Development (supported by AWS Japan)

Developing relatively large-scale Physical AI models centered on VLA using cloud GPU clusters.

Members: 1-3 members per theme
Started: 2026-05

Project Details

This initiative develops relatively large-scale Physical AI models using cloud GPU clusters. We will run multiple development themes including VLA, and plan to release outputs under appropriate licenses with contributor credits.

Objective

Develop and publish practically valuable Physical AI models in a reproducible way.

Theme Proposals

We also welcome topic proposals from KUPAC members.

Implementation Themes

Theme 1: Language-conditioned robot motion generation using motion retargeting

With patented technology and technical guidance from Prof. Yagi, we will build models that generate human-like motions from natural language instructions. The primary target is humanoid robots, and we may consider publishing papers if results are strong.

Theme 2: Feasibility of applying Online/Offline RL to VLA

Using RLinf, we will build RL methods for VLA that perform well in simulation and ultimately target deployment on real hardware. The primary target is dual-arm manipulator robots.

Theme 3: Large-scale GPU-based VLA training insights and parameter-efficient modeling

By training VLA models on large GPU resources, we will accumulate practical insights on training methods, compute resources, stability, and simulation evaluation. We will compare model size, inference speed, memory usage, and task performance to explore parameter-efficient approaches.

Open Roles and Requirements

We welcome both experienced AI robotics contributors and members who want to seriously start in this area.

Open Positions

1-3 members per theme

Schedule

Recruitment: Open for the foreseeable future

Implementation period: 2026/05/01-2026/08/31

Required

  • -Development experience using Python

Preferred

  • -Experience with machine learning frameworks such as PyTorch
  • -Linux development experience
  • -Team development experience

Remote participation is available.

Benefits

  • -We provide compute resources such as GPUs required for development.
  • -Outputs can be listed in your portfolio as achievements.
  • -For substantial outcomes, contributor credits will be shown at publication.